Living Worlds

Research · Populations · Oct 11, 2026

Births with the level set once, on a survey nobody had opened

Met4,528 women, 2013-2015 held out

The test at a glance

What it tests
Accuracy against reality: who has a baby in the next three years, and how many do, with the level set once.
Data
US National Survey of Family Growth (CDC): voice tuned on 2011-2013 and frozen; scored on 4,528 women in 2013-2015.
Held out
The 2013-2015 files, never opened here before registration.
Compared against
Age alone; a regression on the same facts; both fitted on 2011-2013.
Bar
AUC above age alone (95% interval above zero) AND the predicted share inside the observed share's 95% interval.
Result
AUC 0.699 against 0.622: met. Share 0.224 against 0.234 observed (0.222 to 0.246): met.

The births voice, tuned on women's records from 2011-2013 and frozen, was asked about 4,528 women in a survey wave nobody here had opened. It ranked who would have a baby in the next three years well ahead of age alone (AUC 0.699 against 0.622) and slightly ahead of a regression on the same facts (0.685). The earlier test adjusted the level twice for the same fall in births; here it is set once, from age and the national birth rate at each woman's own age, and the voice moves each woman above or below it. Predicted share 0.224 against the observed 0.234 (95% interval 0.222 to 0.246): inside. The earlier rule would have given 0.217, outside. Registered before the data was read; scored once.

Protocol, result and scorer

Next

Births across Africa and South Asia with the Demographic and Health Surveys, and the next NSFG release when it comes out.

Populations

More in this area.